Document not found! Please try again

copyright protection of multimedia documents: from theory ... - CiteSeerX

3 downloads 42589 Views 242KB Size Report
protecting digital still images, audio and video from piracy a matter of urgency. ... A digital signature is an encoded message that matches the content of a particular .... applied to the image under inspection, insert hidden and secret reference ...
COPYRIGHT PROTECTION OF MULTIMEDIA DOCUMENTS: FROM THEORY TO APPLICATION G. Voyatzis, N.Nikolaidis and I.Pitas Department of Informatics, University of Thessaloniki Thessaloniki 54006, Greece Tel,Fax: +3031-996304, e-mail: [email protected] Abstract

Transmission, manipulation and storage of images in digital format is rapidly becoming an everyday practice. Desktop publishing, digital libraries, image databases and the World Wide Web are only some of the application areas that are strongly related to digital imaging technology. The new digital, networked environment necessitates the development of robust and trustworthy encryption, authentication and copyright protection techniques. In this paper we describe a general framework for image copyright protection through image watermarking. In particular we present the main features of an ecient watermarking scheme, discuss robustness issues and describe the three main stages of a watermarking algorithm namely watermark generation, embedding and detection.

1 INTRODUCTION The rapid evolution of digital technology makes the development of reliable and robust schemes for protecting digital still images, audio and video from piracy a matter of urgency. Piracy attacks include illegal access to transmitted data in networks, data content modi cation and production and retransmission of illegitimate copies [1, 2]. The impact of such attacks might be very large both in nancial (pro t losses by unauthorised access and use of data) and security terms. In the following, we clarify three important notions: public key cryptography, authentication and copyright protection.

Public key Cryptography

Data transmitted through network communication lines may be protected from unauthorised receivers by applying techniques based on cryptography [3]. Only persons, who possess the appropriate

Figure 1: Schematic representation of data encryption, authenticity veri cation and watermarking. private key, can decrypt the received data using a public algorithm implemented either in hardware or in software. Furthermore, increase of data size due to encryption should remain within reasonable limits.

Authentication

Data content manipulation can be performed for various legal or illegal purposes (compression, noise removal, malicious data modi cation). The modi ed product is not authentic with respect to the original one. Therefore, users should be able to check the originality of the content of a digital product. Content veri cation can be performed by attaching digital signatures to the transmitted data. A digital signature is an encoded message that matches the content of a particular authentic digital product [3]. Authenticity veri cation procedures are based on public algorithms and public keys. Any \worth noting" modi cation performed in the product or in the signature data should cause veri cation failure.

Copyright Protection

Reproduction of a digital product is easy and inexpensive. In a networked environment (like the Word Wide Web) retransmission of copies all over the world is feasible. Copyright ownership can be violated by persons who illegally claim the product exploitation rights. A copyright protection technique, used by television channels, is the insertion of a visible logo in the digital image. However such a logo can be easily removed or replaced and, subsequently, any evidence about the legal owner is lost. The problem of protecting the intellectual property of digital products has been treated in the last few years with the introduction of the notion of watermarks. Watermarks modify slightly the digital data to embed non-perceptible encoded copyright information. In this paper we discuss watermark e ectiveness in the protection of the intellectual rights on digital products. We will refer, mostly, to watermarking of still digital images. However the concepts introduced in this work can be readily extended to digital audio and video.

2 WATERMARK LITERATURE OVERVIEW A variety of watermarking techniques has been proposed by various authors in the last three years. The proposed algorithms can be classi ed in two main classes on the basis of the utilisation of the original image during the detection phase. Algorithms proposed in [11, 12, 9, 10, 20, 13] do not require the original image whereas in those presented in [17, 18, 16, 7] the original image is input in the detection algorithm along with the watermarked image. Detectors of the second type have the advantage to detect the watermarks in images that have been extensively modi ed in various ways. However detectors of this kind cannot be combined with web-crawling and automatic watermark searching in a digital library. Watermark embedding can be done either in the spatial domain or in an appropriate transform domain (DCT domain [7, 15, 19, 13], Wavelet transform domain [17, 18], Fourier Mellin domain [20], FT domain [19]). In certain algorithms also, the imposed changes take into account the local image characteristics and the properties of the human visual system (perceptual masking) in order to obtain watermarks that are guaranteed to be invisible [15, 12, 13, 17].

3 MAIN FEATURES OF A WATERMARKING SCHEME FOR STILL IMAGES Watermarks are digital signals that are superimposed on a digital image causing alterations to the original data. A particular watermark belongs exclusively to one owner who is the only person that can proceed to a trustworthy detection of its personal watermark and, thus prove the ownership of the host image. Watermarks should possess the following features: Perceptual Invisibility: The modi cations caused by watermark embedding, should not degrade the perceived image quality. However, even hardly visible di erences may become apparent when the original image is directly compared to the watermarked one. We therefore make the assumption that the original product is accessible only to the legal owner and such di erences remain unnoticed by the observer. Complexity: Watermark signals should be characterised by great complexity. This is necessary in order to be able to produce an extensive set of suciently well distinguishable watermarks. An enormous set of watermarks prevents the recovery of a particular watermark by trial and error procedures. In the majority of cases the complexity of a watermark is directly related to the size of the image where it is applied. Associated key: Watermarks should be associated with an identi cation number so called watermark key. The key is used to cast, detect and remove a watermark. Subsequently, the key should be private and characterise exclusively the legal owner. Any digital signal, extracted from a digital image, is assumed to be a valid watermark if and only if it is associated with a key via a well established

algorithm. This condition prevents the creation of counterfeit watermarks discussed extensively by Craver et al [4]. Trustworthy detection: Watermarks should constitute a sucient and trustworthy proof of ownership on a particular product. Detection false alarms should appear very rarely (hopefully never). A particular watermark is a credible evidence for proving copyright ownership when its demonstration in a digital image is followed with insigni cant error probability. Automated detection/search: Watermarks should combine easily with a search procedure that scans any publicly accessible domain in a network environment for illegal deposition of an owner's product. Statistical invisibility: Watermarks should not be recovered using statistical methods. For example the possession of a great number of digital products, watermarked with the same key, should not dispose the watermark by applying statistical methods. Therefore, watermarks should be image dependent. Multiple Watermarking: We should be able to embed a sucient number of di erent watermarks in the same image. Each watermark should be detectable by using the corresponding unique key. This feature seems necessary because we cannot prevent someone from watermarking an already watermarked image. It is also convenient in cases where the copyright property is transferred from one owner to another (a ngerprinting like process [2]). We mention that the legal image owner is the only one that can dispose a copy containing only his/her watermark [5]. Robustness: A digital image can undergo a great deal of di erent modi cations that deliberately (piracy attacks) or not (compression, ltering for noise removal, resizing) a ect the embedded watermark. Obviously, a watermark that is to be used as a means of copyright protection should be detectable up to the point that the host image quality remains within acceptable limits. Because of its importance, the watermark robustness issue will be more thoroughly discussed in section 5.

4 WATERMARKING IMPLEMENTATION FUNCTIONS Generally, watermarking algorithms are described by \private key encode-decode" processing. Although a secure watermarking algorithm based on public key decoding would be very convenient, such an algorithm is very dicult to develop. Let o be the original image of size  . We can de ne as watermark a 2D digital signal of the same size having elements: I

N

M

( ) 2 f?1 0 1g

W i; j

;

;

W

;

0

i < N ;

0

j < M

(1)

A binary or, more generally, a bi-valued form can be also considered. In our de nition, zero values denote image pixels or regions that are not a ected by the watermarking. In a watermarking scheme one can distinguish three fundamental stages: watermark generation, embedding and detection.

4.1 Watermark generation Let W be the set of possible watermark signals. According to the requirement for the existence of an associated key we consider the nite key space K. If I denotes the set of still digital images, a watermark generation procedure should be de ned by the following function :

F :I K!W

;

= F(

W

I; K

)

(2)

where 2 K is the watermark key and 2 I is the image where the watermark will be embedded. For any particular image and a given watermark signal the key extraction should be impossible. F should preferably be a composite function: K

I

I

W

F =T G

;

G :K !W T :W I !W

(3)

;

G may be a non-invertible pseudo-random number generator having seed the input key . T modi es the watermark produced by G to obtain a new watermark 0 according to the image where the watermarking is applied. We remark that the non-invertibility of F is inherited from either G or T . The watermark modi cation function T should take into account only robust image characteristics K

W

W

so that both the original image 0, the watermarked one 0 w ) result in the same watermark : I

I

w

and a modi ed copy of

I

w

(denoted by

I

T(

W; I

0

o ) = T (W; Iw ) = T (W; Iw )

(4)

4.2 Watermark embedding By considering a watermark ( ) produced by F , the embedding process is de ned as a superposition of a 2-D digital signal ( ) onto the original image o ( ). We denote the embedding procedure by E and we de ne it as follows: W i; j

W i; j

I

E :I W  !I R

;

I

w

= E(

o

i; j

I ;W

;)

(5)

l

The real-valued parameter is associated to the embedding watermark energy. Function E can be expressed as a simple watermark superposition : l

w (i; j )

I

=

I

o (i; j )

 ( ) ( )

(6)

L i; j W i; j

where is a two dimensional watermark embedding mask and  denotes a superposition operator including appropriate truncations and quantisation. Coecients ( ) can be either constant for all image pixels or vary according to the local image characteristics. Small values of ( ) may guarantee the watermark invisibility. However, in this case the embedded watermark energy is also small and the watermark can be removed even by image modi cations of moderate strength. The embedding mask should be image speci c and take under consideration the perceiving characteristics of the human visual system. The invertibility of E is desirable but not necessary. L

L i; j

L i; j

L

Figure 2: Watermarking Scheme for casting (top) and detection (bottom). WPA, WEA and WDA denote the algorithms for watermark generation embedding and statistical detection

4.3 Watermark detection Watermark detection is the most important part in a watermarking algorithm. We denote this procedure by the function D. The detector output may be either a binary (yes/no) decision [11, 12] on the existence of a watermark or a longer bitstream carrying various information [8, 9, 10]. When the watermark is image dependent, the associated key 2 K is rst input in F , is created and inserted in D. Note that F should be robust to changes in the image because otherwise it would produce a wrong key when applied on an image that has been manipulated. By taking under consideration the above notions, we de ne the function D : I  K ! f0 1g as follows: K

D( w ) = D( w F ( w I

;W

I

;

I

8 >< 1 if exists )) = > : 0 otherwise W

;K

W

;

(7)

Detection functions of this kind are the most convenient for creating an ecient watermarking framework for copyright protection. Hypothesis testing [12, 10] or watermark similarity correlators [7, 15] can be used as a basis for such detectors. The detector output will form a substantial proving evidence of copyright ownership at a law court provided that it is suciently trustworthy. This requirement suggests that watermark detection should be a publicly known and globally acceptable procedure. By adopting a hypothesis testing framework, possible errors can be classi ed in two categories : Type I error: Watermark is detected although it does not exist in the image. This error is expressed quantitatively by the probability of false alarm ( fa). Type II error : Watermark is not detected in the image although it exists. Thus, we get an error probability of watermark rejection ( rej ). The total error probability is err = fa + rej and the detection performance increases when err decreases. However, the reliability of the detection is associated exclusively with the false alarm probability and the following (almost linear) convergence should be satis ed for a randomly selected watermark: lim [ fD( W ) = 1g] = 0 (8) C (K)!1 P

P

P

P

P

P rob

P

I

;W

where (K) denotes the cardinality of the key set. We should mention here that the two types of error compete each other; by decreasing fa, rej increases and vice versa. C

P

P

5 ROBUSTNESS ISSUES A watermark that is of some practical use should be robust to image modi cations up to a certain amount. The most common image manipulations are the following: Geometric distortions (scaling, cropping, deletion or insertion of lines/columns, re ection) To cope with such distortions one can embed the watermark in the Fourier Mellin Coecient space [20], perform a search within the space of all possible geometric distortions to nd the one that has been applied to the image under inspection, insert hidden and secret reference marks, or use correlation based techniques. Compression Compression algorithms tend to remove visually insigni cant information which is usually where watermarks reside. Some authors propose placing the watermark in the perceptually important components of the image or use watermarks with lowpass characteristics. Filtering Attacks based on lowpass ltering (mean, median) can be treated using watermark signals having lowpass characteristics. Color Quantisation/ Color-Brightness modi cations (Histogram modi cation/equalisation). A watermark scheme that copes successfully with all possible attacks has not been proposed yet.

References [1] B.Macq and J.J.Quisquater, \Cryptology for Digital TV Broadcasting", Proceeding of the IEEE, vol 83, no 6, pp.944-957, 1995. [2] J.F.Delaigle, J.M.Boucqueau, J.J.Quisquater and B.Macq, \Digital images protection techniques in a broadcast framework : an overview", Proceedings of ECMAST'96, vol.2, pp 711-727, 1996. [3] D.R.Stinson, \Cryptography, theory and practice", CRC Press, 1995. [4] S.Craver, N.Memon, B.L. Yeo and M.M.Yeung, \Can invisible watermarks resolve rightful ownerships?", IBM Tech.Report RC20509, 1996. [5] J.Zhao, "A WWW service to embed and prove digital copyright watermarks", Proceedings of ECMAST'96, vol.2, pp 695-709, 1996. [6] I.J.Cox and M.L.Miller \A review of watermarking and the importance of perceptual modeling", Proc. of Electronic Imaging'97, February 1997. [7] I.J.Cox, J.Kilian, T.Leighton and T.Shammoon \Secure Spread spectrum Watermarking for Multimedia", Tech. Report, NEC research Institute, no 95-10, 1995.

[8] C.T.Hsu and J.L.Wu, \Hidden signatures in Images" Proceedings of ICIP'96, vol III, pp 223-226, 1996. [9] G.Voyatzis and I.Pitas, \Applications of Toral automorphisms in image watermarking", Proceedings of ICIP'96, vol II, pp. 237-240, 1996. [10] G. Voyatzis and I. Pitas, \Embedding Robust Logo Watermarks in Digital Images", Proceedings of DSP'97, vol 1, pp. 213-216, 1997. [11] I.Pitas, \A method for signature casting on digital images", Proceedings of ICIP'96, vol III, pp 215-218, 1996. [12] N.Nikolaidis and I.Pitas \Robust image watermarking in the spatial domain", Signal Proc. special issue on Copyright Protection and Access control, vol 66, no 3, pp. 385-403, 1998. [13] A. Piva, M. Barni, F. Bartolini, V. Cappellini, "DCT-based watermark recovering without resorting to the uncorrupted original image", Proceedings of ICIP'97, Santa Barbara, CA, USA, October 26-29, 1997, Vol I, pp. 520-523. [14] R. B. Wolfgang and E. J. Delp, "A Watermark for Digital Images," Proceedings of ICIP'96, September 16-19, 1996, Lausanne, Switzerland, pp. 219-222. [15] Mitchell D. Swanson , Bin Zhu , and Ahmed H. Tew k , "Transparent Robust Image Watermarking," Proc. of the 1996 IEEE Int. Conf. on Image Processing , Vol. III, PP. 211-214, 1996. [16] J. O'Ruanaidh, W. Dowling, F. Boland, \Watermarking digital images for copyright protection", IEE Proceedings on Vision, Image and Signal Processing, 143(4), pp 250-256, August 1996. [17] D. Kundur, D. Hatzinakos, \A Robust Digital Image Watermarking Method using WaveletBased Fusion", Proceedings of ICIP'97, Santa Barbara, CA, USA, October 26-29, 1997, Vol I, pp. 544-547. [18] X.-G. Xia , C. G. Boncelet, and G. R. Arce, \ A Multiresolution Watermark for Digital Images" Proceedings of ICIP'97, Santa Barbara, CA, USA, October 26-29, 1997, Vol I, pp. 548-551. [19] J. O'Ruanaidh, W. Dowling, F. Boland, \Phase watermarking of digital images", Proc. 1996 IEEE Int. Conference on Image Processing (ICIP 96), vol III, pp 239-242. [20] J. O'Ruanaidh,, T. Pun, \Rotation, Scale and Translation Invariant Digital Image Watermarking", Proceedings of ICIP'97, Santa Barbara, CA, USA, October 26-29, 1997, Vol I, pp. 536-539.